SupplementMagnetic resonance virtual histology for embryos: 3D atlases for automated high-throughput phenotyping
Research Highlights
►Using MRI histology, we identified pituitary phenotypes in Hesx1 mutant embryos. ►An atlas of CD-1 embryos, gave an average image with enhanced anatomical detail. ►This enabled the rapid assessment of morphological differences between embryo groups. ►Mean brain volume was reduced in Chd7+/- compared to Chd7+/+ mice (39 vs. 42 mm3). ►Body, olfactory, pituitary volume differences seen between CD-1 and Chd7+/+ mice.
Introduction
In the wake of the first draft of the full mouse genome sequence (Mouse Genome Sequencing Consortium), large-scale mutagenesis programmes are underway (International Mouse Knockout Consortium) that will produce mice with gene knockouts for each of the approximately 25,000 genes in the mouse genome. Analysis of these mice in the coming years will give new insights into the genetic basis of human disease, as novel genes are identified that impact upon mammalian physiology and morphology. Mouse embryos in particular may be studied to determine the role of genes on development and congenital abnormalities. With an increasing number of new mutants, effective methods of identifying novel phenotypes in these embryos will be crucial.
Current phenotyping of embryo morphology is generally achieved by histological examination using microscopy. Specimens are dehydrated, wax embedded and thinly sectioned (2–8 μm) (Kaufman, 1992), providing high resolution 2D data and tissue sections that may also be stained for gene and protein expressions. Episcopic imaging is a development of this process, where autofluorescence of each tissue slice can be photographed and combined to generate high resolution 3D volume datasets (1–2 μm isotropic resolution) (Weninger and Mohun, 2002). However these histological approaches are time-consuming, introduce distortions into the final 3D image due to the sectioning process and do not readily enable rapid screening as only one embryo may be imaged at a time.
MRI is now an established method for non-invasive embryo imaging, beginning with the early work of Smith et al. (1994). High resolution 3D datasets with isotropic resolutions of down to 12 μm are created (Smith et al., 1996) with excellent soft tissue contrast, allowing the visualisation and segmentation of individual organ structures (Dhenain et al., 2001). MRI is also capable of high-throughput screening of multiple ex-vivo embryos (up to 32 in a single overnight scan) with the combination of a large volume imaging coil (Schneider et al., 2004) and fixation in an MR contrast agent (Cleary et al., 2009). Diffusion tensor imaging (DTI) — an alternative MRI technique — has also been used to investigate the structure of the embryo CNS by exploring the degree and direction that water is able to diffuse along neuronal axons (Zhang et al., 2003). Although this technique is ideal for investigating white matter, it is impractical for high-throughput imaging, as many hours are needed to generate connectivity maps in a single brain.
Other imaging methods such as optical projection tomography (OPT) and micro-computed tomography (μCT) are also able to non-destructively produce 3D datasets. OPT can create images of embryos that combine both anatomical structure and gene expression with conventional fluorophores, at high resolution (5–10 μm) (Johnson et al., 2006, Sharpe, 2004). However the technique requires embryos which are partially transparent thereby making its use challenging in older subjects (> 13.5 days post coitum, dpc) (Schneider and Bhattacharya, 2004). μCT is also capable of acquiring high resolution datasets (typically less than 27 μm) in a short scan time (~ 2 h) (Johnson et al., 2006). Although conventionally μCT has difficulty in distinguishing soft tissues, which have inherently low contrast due to a narrow range of CT numbers (Holdsworth and Thornton, 2002), the use of CT contrast agents as tissue stains, such as osmium tetroxide (Johnson et al., 2006) and potassium triiodide (Degenhardt et al., 2010), have improved its ability to discriminate tissues. μCT is also particularly suited to skeletal studies, as it can produce excellent images of dense radiopaque structures such as bone (Oest et al., 2008).
Despite the availability of these advanced imaging techniques, any embryo dataset must still be manually assessed through inspection by a trained observer. As high-throughput analysis is increasingly demanded, conventional visual assessment for abnormalities is likely to become labour-intensive and insensitive.
Advanced computational techniques such as segmentation–propagation and voxel-based morphometry (VBM) have been used to investigate populations in both clinical and adult mouse MRI studies (Ashburner and Friston, 2000, Calmon and Roberts, 2000, Lerch et al., 2008, Sawiak et al., 2009). These methods enable anatomical differences between groups to be identified with little manual intervention or visual assessment. Integral to these techniques is the use of an atlas, a spatial average image of the whole population (Ashburner and Friston, 2000, Calmon and Roberts, 2000) created by finely warping individual subject images together to locally align anatomical features. Segmentation–propagation is a quantitative method for making volumetric measurements. After segmenting a volume of interest on the atlas image, such as the brain or heart, differences may then be identified between groups in the population by propagating the segmented volume to all individuals, thus providing the group mean and standard deviation from a single volume of interest on the atlas image. However, while average images of registered wild-type embryos have been reported previously (Zamyadi et al., 2008), there has so far been no application of atlas methods for phenotypic assessment. We envisage that a combination of multiple whole-embryo imaging with image processing techniques would allow the creation of an average embryo atlas from a population and enable automated phenotypic comparisons between transgenic and wild-type littermates.
In this study there were three stages to the investigation of our embryo atlas. We started by addressing the lack of brain tissue contrast on MR images. Initially we developed a contrast enhanced MR technique to produce structural detail in the embryo CNS. This was assessed in Hesx1-/- and Hesx1I26T/I26T mice, models of septo-optic dysplasia (SOD) (Dattani et al., 1998, Sajedi et al., 2008). Combining our contrast enhanced protocol with computational methods, we generated an MRI atlas for a population of CD-1 embryos and compared this against histology. Finally, this enabled the use of a segmentation–propagation technique to assess brain and cardiac phenotypic differences between CD-1, C57BL/6 strains and Chd7+/- knockout mice (a model of the condition CHARGE syndrome) (Bosman et al., 2005, Randall et al., 2009) based on a novel population atlas.
Section snippets
Animal preparation
Pregnant female mice were sacrificed by cervical dislocation. Embryos were then dissected from the mother and transferred to warm Hanks solution. Their umbilical cords were cut and the embryos were allowed to bleed out into the solution. The embryos were then fixed in a solution of 4% formaldehyde and Magnevist MR contrast (Gadolinium-DTPA, Bayer-Schering Pharma, Newbury, UK) and left on a rotator. The embryos were removed and embedded in a 50 ml centrifuge tube using 1% agarose gel doped with
Micro-MRI for the investigation of embryo brain anatomy in 3D
15.5 dpc CD-1 embryos were fixed in 4% formaldehyde solution with a gadolinium–chelate contrast agent (Gd-DTPA) and imaged at 52 μm resolution, interpolated to 26 μm by zero-filling. Obtaining tissue contrast, particularly in brain structures is often difficult in embryo imaging. We have evaluated images of the embryo CNS acquired from specimens fixed in four contrast agent concentrations (2 to 16 mM) and two fixation durations (three days and two weeks) to determine optimal preparation. Two weeks
Discussion
The main findings of this study were: i) development of embryo-specific MR methods and contrast agent protocols enabled high resolution structural imaging with enhanced anatomical detail in the CNS, ii) using these methods, we have identified the structural consequences in Hesx1-/- and Hesx1I26T/I26T mouse mutants, iii) we have generated a composite whole-embryo atlas using computational methods with excellent anatomical tissue contrast, and finally iv) we presented a novel application of a
Conclusion
We have presented an average atlas generated from high-throughput multi-embryo MRI datasets. We have demonstrated that anatomy is conserved in wild-type CD-1 embryos, enabling the creation of an average embryo atlas that has effectively enhanced anatomical detail. Volumetric analysis of three groups of embryos using segmentation–propagation found significant brain volume differences between CD-1, Chd7+/- knockout and wild-type. Significant differences were observed in whole-body and olfactory
Acknowledgments
We are grateful to Dr. Karen McCue, Sarah Beddow and Vanessa Kyriakopoulou, Molecular Medicine Unit, UCL Institute of Child Health, for assistance with Chd7 and CD-1 embryo preparation. Also Sandra De Castro, Neural Development Unit, UCL Institute of Child Health, for performing CD-1 histology. We thank Dr. Gerard Ridgway, Prof. David Gadian and Prof. Elizabeth Fisher for helpful suggestions, and Tristan Clark for assistance with the UCL Department of Computer Science computer cluster. We
References (48)
- et al.
Voxel-based morphometry—the methods
Neuroimage
(2000) - et al.
Automatic measurement of changes in brain volume on consecutive 3D MR images by segmentation propagation
Magn. Reson. Imaging
(2000) - et al.
Neuroanatomical differences between mouse strains as shown by high-resolution 3D MRI
Neuroimage
(2006) - et al.
High resolution three-dimensional brain atlas using an average magnetic resonance image of 40 adult C57Bl/6J mice
Neuroimage
(2008) - et al.
Micro-CT in small animal and specimen imaging
Trends Biotechnol.
(2002) - et al.
Towards a microMRI atlas of mouse development
Comput. Med. Imaging Graph.
(1999) - et al.
Automated deformation analysis in the YAC128 Huntington disease mouse model
Neuroimage
(2008) - et al.
A three-dimensional digital atlas database of the adult C57BL/6J mouse brain by magnetic resonance microscopy
Neuroscience
(2005) - et al.
T2 relaxation time as a marker of brain myelination: experimental MR study in two neonatal animal models
J. Neurosci. Meth.
(1997) - et al.
Fast free-form deformation using graphics processing units
Comput. Meth. Programs Biomed.
(2010)